sign in sign up
<<
Home , Lists of planets, Low Mass, Planet

The Astrophysical Journal, 718:1186–1199, 2010 August 1 doi:10.1088/0004-637X/718/2/1186 C 2010. The American Astronomical Society. All rights reserved. Printed in the U.S.A.

DISCOVERY OF A LOW- COMPANION TO A -RICH F WITH THE MARVELS PILOT PROJECT

Scott W. Fleming1,JianGe1, Suvrath Mahadevan1,2,3, Brian Lee1, Jason D. Eastman4, Robert J. Siverd4, B. Scott Gaudi4, Andrzej Niedzielski5, Thirupathi Sivarani6, Keivan G. Stassun7,8, Alex Wolszczan2,3, Rory Barnes9, Bruce Gary7, Duy Cuong Nguyen1, Robert C. Morehead1, Xiaoke Wan1, Bo Zhao1, Jian Liu1, Pengcheng Guo1, Stephen R. Kane1,10, Julian C. van Eyken1,10, Nathan M. De Lee1, Justin R. Crepp1,11, Alaina C. Shelden1,12, Chris Laws9, John P. Wisniewski9, Donald P. Schneider2,3, Joshua Pepper7, Stephanie A. Snedden12, Kaike Pan12, Dmitry Bizyaev12, Howard Brewington12, Olena Malanushenko12, Viktor Malanushenko12, Daniel Oravetz12, Audrey Simmons12, and Shannon Watters12,13 1 Department of , University of Florida, 211 Bryant Center, Gainesville, FL 326711-2055, USA; scfl[email protected]fl.edu 2 Department of Astronomy and Astrophysics, The Pennsylvania State University, 525 Davey Laboratory, University Park, PA 16802, USA 3 Center for and Habitable Worlds, The Pennsylvania State University, University Park, PA 16802, USA 4 Department of Astronomy, The Ohio State University, 140 West 18th Avenue, Columbus, OH 43210, USA 5 Torun´ Center for Astronomy, University, ul. Gagarina 11, 87-100, Torun,´ Poland 6 Indian Institute of Astrophysics, Bangalore 560034, 7 Department of and Astronomy, Vanderbilt University, Nashville, TN 37235, USA 8 Department of Physics, Fisk University, 1000 17th Ave. N., Nashville, TN 37208, USA 9 Department of Astronomy, University of Washington, P.O. Box 351580, Seattle, WA 98195, USA 10 NASA Science Institute, Caltech, MS 100-22, 770 South Wilson Avenue, Pasadena, CA 91125, USA 11 Department of Astronomy, California Institute of Technology, 1200 E. California Blvd., Pasadena, CA 91125, USA 12 Apache Point Observatory, P.O. Box 59, , NM 88349-0059, USA 13 Institute for Astronomy, 34 Ohia Ku St., Pukalani, HI 96768-8288, USA Received 2010 April 26; accepted 2010 June 8; published 2010 July 13

ABSTRACT We report the discovery of a low-mass companion orbiting the metal-rich, main F star TYC 2949-00557-1 during the Multi-object APO Exoplanet Large-area Survey (MARVELS) pilot project. The host star has an Teff = 6135 ± 40 K, logg = 4.4 ± 0.1, and [Fe/H] = 0.32 ± 0.01, indicating a mass of M = 1.25 ± 0.09 M and R = 1.15 ± 0.15 R. The companion has an of 5.69449 ± 0.00023 days and straddles the burning limit with a of 64 MJ, and thus may be an example of the rare class of brown dwarfs orbiting at distances comparable to those of “Hot .” We present relative that demonstrates that the host star is photometrically stable at the few millimagnitude level on time scales of hours to , and rules out transits for a companion of radius  0.8 RJ at the 95% confidence level. Tidal analysis of the system suggests that the star and companion are likely in a double synchronous state where both rotational and orbital synchronization have been achieved. This is the first low-mass companion detected with a multi-object, dispersed, fixed-delay interferometer. Key words: brown dwarfs – planetary systems – : low-mass Online-only material: color figures

   1. INTRODUCTION (12 MJ m sin i 80 MJ ) at separations of a 5AU, relative to more massive stellar companions and less massive Studies of the frequency, parameter distributions, and corre- planetary companions (Marcy & Butler 2000). Note that we lations of extrasolar require homogeneous samples of denote i as the inclination angle between the companion’s hundreds of planets to obtain statistically significant results. orbital vector and the line of sight, and Moreover, such a sample have well-understood complete- we reserve I as the inclination angle of the stellar ness limits, selection effects, and biases, which are easiest to ob- axis to the line of sight. While the frequency of tain from a single, large-scale survey. Given current constraints companions at larger separations is still relatively uncertain on the frequency of giant planets, detection of such a large sam- (e.g., Metchev & Hillenbrand 2009), a meta-analysis of sets of ple of planetary systems generally requires a precision radial known companions to solar-type stars by Grether & Lineweaver velocity (RV) survey of many thousands of stars. Such a survey (2006), with corrections for observational bias, confirmed the also provides a wealth of ancillary science. In particular, it is lack of brown dwarfs at close separations. These authors place exquisitely sensitive to more massive companions, and because the “driest” part of the brown dwarf desert at ∼20–50 MJ , with it targets a large and broad sample of host stars, it is naturally a frequency of companions  0.5% in this range of masses. sensitive to rare binary systems in poorly explored regions of Although there has been a steady increase in the number of parameter space. known brown dwarf candidates via the RV technique (Marcy Of particular interest are the constraints on the frequency et al. 2001;Udryetal.2002; Endl et al. 2004; Patel et al. and parameter distributions of low-mass companions to solar- 2007; Wittenmyer et al. 2009; Kane et al. 2009; Jenkins et al. type stars with masses near the hydrogen burning limit. One of 2009; Niedzielski et al. 2009; Omiya et al. 2009), most of the early results from precise RV searches was the apparent these detections have been at separations a  0.8 AU. Notable paucity of brown dwarf companions with minimum masses exceptions include the transiting brown dwarf CoRoT-Exo-3b

1186 No. 2, 2010 DISCOVERY OF TYC 2949-00557-1b 1187 with a period of ∼4 days orbiting an F3V star (Deleuil et al. Table 1 2008), and HD41004Bb with a period of ∼1 orbiting the MPP RV Observations a M dwarf component of a K–M (Santos et al. BJDTDB RV σRV 2002). Brown dwarfs at such short orbital separations are (ms−1) (ms−1) of particular interest for several reasons. First, the frequency 2454101.69079 13339 94 of such systems as a function of their physical and orbital 2454105.75520 19819 98 parameters provide diagnostics that may be able to distinguish 2454106.70228 14981 96 between the various mechanisms that have been invoked for their 2454128.62211 19018 143 formation and dynamical (e.g., Armitage & Bonnell 2454128.86491 17170 91 2002; Matzner & Levin 2005). In particular, these systems offer 2454130.83623 13299 94 observational constraints on the poorly understood theory of 2454136.64529 15043 85 tidal interactions between host stars and close companions (e.g., 2454136.85894 15854 139 2454163.72245 14882 91 Mazeh 2008; Pont 2009). Second, these systems are much more 2454188.69749 21003 136 likely to than their longer-period counterparts, as the 2454191.68122 15421 97 transit probability is inversely proportional to orbital separation. 2454194.69216 21197 115 Transiting systems yield valuable measurements on the masses, 2454195.68446 24453 102 radii, and mean densities of brown dwarfs (Stassun et al. 2006, 2454217.61115 22758 79 2007; Deleuil et al. 2008). Here we report the discovery of a candidate short-period, Note. a Errors are not scaled to account for systematics. brown dwarf companion to the metal-rich star TYC 2949- 00557-1, a F star with apparent brightness ter; previous observations with a single-object DFDI instrument ∼ V 12.1. This companion was discovered as part of the at Kitt Peak National Observatory (KPNO) resulted in the first Multi-object APO Radial Velocity Exoplanet Large-area Survey extrasolar discovered via this technique (Ge et al. 2006b), (MARVELS) pilot project (hereafter MPP). The MPP used the as well as the first confirmed planet via DFDI (van Eyken et al. W. M. Keck Exoplanet Tracker (Keck ET) instrument (Ge et al. 2004) and the ability to measure precise, absolute RVs with 2006a) on the Sloan Digital Survey (SDSS) 2.5 m DFDI (Mahadevan et al. 2008). (Gunn et al. 2006) at the Apache Point Observatory. The Keck ET instrument is a multi-object (59 targets per exposure), 2. DOPPLER OBSERVATIONS dispersed fixed-delay interferometer (DFDI; Ge et al. 2002;Ge 2002; Erskine 2002; Erskine et al. 2003). In this instrument, 2.1. MPP Observations fiber-fed from the telescope is first passed through an iodine cell that acts as a stable wavelength reference. This The MPP targeted 708 F, G, and K dwarfs with 7.6

Table 2 KPNO/ET RV Observations

a BJDTDB RV σRV (ms−1) (ms−1) 2454749.97070 266 58 2454751.88336 1316 55 2454751.94579 1985 50 2454752.89657 7580 54 2454752.95821 8026 58 2454753.90161 9651 77 2454753.96355 9548 65 2454754.90388 4739 53 2454754.96580 4502 52 2454755.91284 −694 47 2454755.97519 −996 52

Note. a Errors are not scaled to account for systematics.

produces a DFDI spectrum from which RVs are measured. The two beams are processed separately, and their measured RVs are Figure 1. Lomb–Scargle periodogram for the MPP data (14 epochs). A clear and highly significant peak at P = 5.68 days can be seen, with a power of ∼217, combined via a weighted average based on the RV uncertain- which has a false alarm probability based on scrambling the data of < 0.01%. ties. Table 2 contains the dates and velocities for the KPNO ET Powers corresponding to false alarm probabilities of 1%, 0.1%, and 0.01% are measurements. Unlike the results from the MPP, the velocities also shown. presented in Table 2 are relative RVs, i.e., measured relative to one of the epochs. Note that none of the velocities are exactly The resulting power spectrum is shown in Figure 1, revealing a zero because an instrumental drift has been subtracted off based clear peak at P = 5.68 days with a power of ∼217. To assess on the calibration lamps, and they are zeroed to a different epoch the significance of this peak, we ran a Monte Carlo simulation than the star. Because these data are relative RVs, an offset exists with 105 trials. For each trial, we scrambled the times of the between the values in Tables 1 and 2 that must be included as data points, computed the periodogram, and recorded the most an additional parameter when performing a combined analysis significant peak. We found no trials with power greater than that of the two data sets. of actual data, indicating a false alarm probability of < 10−5. The advantage of a multi-object instrument is that the multiple 2.3. HET Doppler Observations targets that are observed simultaneously can be used to check Observations of the candidate were also conducted using the for systematic trends in data. Thus, as an additional R = 60,000 mode of the high resolution spectrograph (HRS) check on the reliability of the observations, we constructed (Tull 1998) on the Hobby–Eberly Telescope (HET) telescope periodograms for the other 58 objects on the plate and calculated (Ramsey et al. 1998) in the queue scheduled mode (Shetrone the power for each object at the period of the suspected et al. 2007). The spectra consisted of 46 echelle orders recorded companion. Any common systematics present in the data due to on the “blue” CCD detector (407.6–592 nm) and 24 orders on the sampling rate or instrumental effects will result in significant the “red” one (602–783.8 nm). The spectral data used for RV power at a common period for other targets. None of the measurements were extracted from the 17 orders that cover the other 58 targets have significant power at the period of TYC 505–592 nm range of the iodine spectrum. A total of 10 Doppler 2949-00557-1; the next strongest candidate has a false alarm RV measurements were obtained spanning 83 days from 2008 probability (FAP) at that period of 77% and the other targets December through 2009 February. The starlight was passed have FAPs > 99%. The best-fit amplitude derived from the through an iodine cell to provide a stable reference to calibrate −1 periodogram is ∼5500 m s , indicating a minimum mass in the instrument drift. Two exposures without the iodine cell were brown dwarf regime for a solar-type primary. Given that brown taken to act as stellar templates. Due to the faintness of the dwarf companions in this period range are rare, we decided target, the RVs were computed relative to each template and to obtain additional precise RV measurements, high-resolution a mean value was determined. The results for both templates spectra, absolute photometry, and precise relative photometry agree to within 3σ . Table 3 contains the dates and velocities for time series, in order to better characterize the primary and the HET measurements. Similar to the results from the KPNO ascertain the nature of the companion. ET in Table 2, these velocities are relative RVs, and therefore an offset exists between these values and the ones in both Tables 1 2.2. KPNO Doppler Observations and 2. Observations for the purpose of confirming the Doppler vari- 2.4. Combined RV Analysis ability and were conducted with the Exoplanet Tracker (ET) instrument (Ge et al. 2006b) at the KPNO with the 2.1 m In order to check for consistency, we first fit the MPP, KPNO, telescope. Two observations separated by several hours were and HET data sets individually to a seven parameter RV fit, taken each night over seven consecutive nights starting on 2008 where the seven parameters are the velocity semi-amplitude October 10. A total of 11 usable epochs were obtained. Integra- K, eccentricity e, argument of periastron ω, period P, time of tions consisted of 60 min exposures bracketed by exposures of a inferior of the companion Tc, velocity zero point γ , Tungsten lamp passing through an iodine gas cell that acts as a and linear slope γ˙ (in order to allow for additional companions calibration for instrument drift. Each arm of the interferometer or systematic drifts). The best-fit solution was found using a No. 2, 2010 DISCOVERY OF TYC 2949-00557-1b 1189

Table 3 the errors to χ 2/dof = 1. Specifically, we investigated HET/HRS RV Observations four different cases for treatment of the MPP (KPNO data come a BJDTDB RV σRV from a similar pipeline) and HET RV uncertainties: a scaling (ms−1) (ms−1) of the errors by a constant factor, an addition in quadrature of 2454807.74076 0 69 a constant error, a removal of suspected outliers based on the 2454808.98051 4674 67 of the RV uncertainty followed by scaling, and a 2454825.69507 2469 107 treatment of all data points with a constant error value. Ten 2454829.70100 2482 66 Markov Chain Monte Carlo (MCMC) simulations were run for 2454881.75734 −23 286 each case. The starting values for the parameters in these chains 2454882.78369 3178 51 were chosen to span a range that is large with respect to the 2454883.76208 9451 63 expected 1σ uncertainty, the chains were stopped after reaching 2454887.75592 165 48 convergence as defined in Ford (2006), and then the chains were 2454889.74685 10881 51 2454890.75523 11513 47 merged. Analyzing the MPP and HET data separately yields discrepant ∼ Note. a Errors are not scaled to account for systematics. periods at the 2σ level for all cases of error treatment. We also found that the choice of error treatment can affect the derived value of e cos ω from the MPP data as well at the ∼ hybrid downhill-simplex fit to the nonlinear parameters and an 1σ level. However, the other parameters from the MPP data, exact (linear) fit to the linear parameters. There are 14 MPP as well as all parameter values from the HET data, were not points, 11 KPNO points, and 10 HET points, so there are seven, significantly affected by different treatments of the uncertainties. four, and three degrees of freedom (dof), respectively. From this test, we conclude that there is no strong justification For the MPP fit, we find a χ 2/dof of 44. Given that the data for removing any data points from the fit, so we conducted the points basically follow the model, and that the more precise final joint analysis of all three data sets where each set of errors HET data (whose error bars are overestimated, see below) fit the is scaled by a constant factor. model well, the large χ 2/dof indicates that there are systematic The MPP errors are scaled by a factor of 6.64, the KPNO uncertainties in addition to the photon noise, and thus the errors errors are scaled by 2.64, and the HET errors are scaled by a 2 ∼ are severely underestimated. Given the large median QF = 3.25 factor of 0.3 such that the reduced χ is 1 when each is fit found for the majority of the (likely constant) stars in this field, independently. The fit including eccentricity as a free parameter this result is not surprising. Indeed, there is a known systematic is consistent with zero eccentricity, therefore we run a second = error in DFDI when utilizing an iodine cell in the stellar beam fit that e 0. In that case, the error scalings are factors { } path (van Eyken et al. 2010). For the KPNO fit we find a of 6.82, 2.71, 0.24 for MPP, KPNO, and HET, respectively. χ 2/dof of 6.96, once again indicating that the uncertainties We further scale the errors of all three data sets by a factor of are underestimated. For the HET fit, we find a χ 2/dof of 0.09, 1.30 for the case where eccentricity is left as a free parameter, therefore the uncertainties are likely overestimated for the HET and 1.25 for the case where eccentricity is fixed to zero. This 2 = data set. It is worth noting that a statistically significant slope is scaling is done so that the χ /dof 1 in the combined fit, and is found when fitting the HET data. Fitting the HET data without necessary due to the systematics present in the data sets. Given a slope produces a significantly worse χ 2/dof. the close separation of the companion, it is expected that the We performed additional RV fitting to test the significance orbit has been tidally circularized, consistent with our findings. of the HET slope. Fitting the HET data with no slope and We therefore treat the case with eccentricity fixed at zero as our eccentricity forced to zero still results in a χ 2/dof  1. Since final values, but quote the parameters from both cases in Table 4, this is the simplest model of an orbiting companion, it confirms which contains the values of the orbital parameters for the case that the HET uncertainties are overestimated. Several different of non-zero eccentricity (eccentric) and eccentricity fixed at 0 models, in which slope is a free parameter, eccentricity is a (circular). free parameter, or both are free parameters, all result in lower Figure 2 shows the final results of the joint RV fitting and χ 2/dof. There is no evidence of non-zero eccentricity in any of fixing the eccentricity at 0. MPP data are the blue squares, the best-fit solutions, but a significant slope is found in all cases KPNO data are the green triangles, HET data are the red when left as a free parameter. To be self-consistent, we allow circles, and the systemic velocity γ0 has been removed. We = ± −1 for slopes in the MPP and KPNO data fitting as well, and note find γ0 18.68 0.24 km s for the star, with an offset ± −1 that their best-fit slopes are consistent with the HET value, but between the MPP and KPNO data of 14.90 0.25 km s and an ± −1 are much more poorly constrained due to the much larger RV offset between the MPP and HET data of 12.61 0.24 km s . ± uncertainties in those data sets. As a final check, we fit all data The final orbital period is determined to be 5.69449 0.00023 ± −1 sets with eccentricity forced to zero and no slope, and find that days and an RV semi-amplitude of 6.113 0.009 km s .We the other orbital parameters are not qualitatively different from searched for an additional signal in the residuals from the joint fit the result where slope is left as a free parameter. We therefore that might be caused by an additional companion in the system, choose the case of zero eccentricity and non-zero slope as our but found no other frequencies with significant power. preferred solution. Uncertainties in the fitting parameters will be inaccurate if 3. RELATIVE TIME SERIES PHOTOMETRY it is determined using misestimated RV errors. It is therefore OF THE HOST STAR important to attempt to correct the errors such that χ 2/dof ∼ 1. However, given that we do not know why the errors are Photometric observations are an important step in analyzing misestimated, particularly in the case of underestimated errors, low-mass companions discovered via the Doppler technique. the appropriate method to correct the uncertainties is not clear. High-precision photometry can be used to search for transits Our approach was to try several different ways of correcting of the companion. Additionally, time-series photometry can be 1190 FLEMING ET AL. Vol. 718

Figure 2. Results from the combined MPP, KPNO, and HET analysis, where the MPP errors are scaled by a factor of 8.49, KPNO errors are scaled by a factor of 3.37, and the HET errors are scaled by a factor of 0.3. The eccentricity is fixed at e = 0. MPP data are in blue, KPNO data are in green, and HET data are in red. The systemic velocity γ0 has been removed. (A color version of this figure is available in the online journal.)

Table 4 as part of the Kilodegree Extremely Little Telescope (KELT) Best-fit Dynamical Properties of TYC 2949-00557-1 transit survey. Neither data sets show any evidence for Parameter Value Uncertainty Value Uncertainty variability of the host star.

Eccentric Case Circular Case 3.1. Relative Photometry from Hereford Arizona Observatory Period (days) 5.69459 0.00029 5.69449 0.00023 K(kms−1) 6.109 0.014 6.113 0.009 Initial photometric observations of the primary were per- Tc (BJDTDB-2454000) 868.9878 0.0042 868.9877 0.0016 formed on four nights in 2009 (February 19, 21, and 27, and e cos ω 0.0000 0.0015 0. ... 16) at the Hereford Arizona Observatory (observatory − +0.0011 e sin ω 0.0005 −0.0019 0. ... code G95 in the IAU Center), a private facility in +0.0019 e 0.0017 −0.0017 0. ... southern Arizona. Additional observations were made in 2010 ω π ... (rad) 4.69 1.58 2 on April 15 to search for transits based on an updated transit mmini (MJ ) 64.3 3.0 64.3 3.0 −1 from the combined RV analysis (Section 2.4). All Systemic velocity γ0 (km s ) 18.67 0.26 18.68 0.24 −1 data were taken with an 11 inch Celestron Schmidt-Cassegrain KPNO offset (γ0 − γkpno,kms ) 14.89 0.26 14.90 0.25 −1 (model CPC 1100) telescope that is fork-mounted on an equato- HET offset (γ0 − γhet,kms ) 12.59 0.26 12.61 0.24 −1 −1 rial wedge, an SBIG ST-8XE CCD with a KAF 1602E detector, MPP slope γ˙mpp (km s day ) −0.0044 0.0065 −0.0037 0.0065 −1 −1 KPNO slope γ˙kpno (km s day ) 0.016 0.034 0.013 0.033 and an SBIG AO-7 tip-tilt image stabilizer used to maintain −1 −1 HET slope γ˙het (km s day ) −0.0010 0.00035 −0.0011 0.00024 the field at a fixed position on the CCD. The observations in Total σRV Scale Factor (MPP) 8.64 ... 8.49 ... 2009 were done without a filter (“C” band), resulting in an ef- Total σRV Scale Factor (KPNO) 3.43 ... 3.37 ... fective central wavelength of ∼570 nm between Johnson V and  Total σRV Scale Factor (HET) 0.39 ... 0.30 ... R bands. The observations in 2010 were taken with a Sloan r Combined χ 2/dof 1.69 ... 1.55 ... filter. Data toward the end of the night on 2010 April 15 were taken at very high air mass (out to sec z = 5.7), resulting is somewhat degraded photometric precision. used to rule out stellar mechanisms of Doppler variability, such Figure 3 shows the relative photometry over the five nights, as chromospheric activity due to or stellar pulsations. which demonstrate that the primary star is intrinsically stable on In the case of stars with detectable starspots, time-series pho- the time scale of several hours, at the level of 2–4 millimagni- tometry can be used to determine a rate. In this tudes. Based on the final ephemerides determined in Section 2.4, section, we present and analyze time series relative photometry only the last night (2010 April 15, top row) covers possible of TYC-2949-00557-1 from two sources: relatively precise (few times of predicted transits. The vertical bars are the predicted mmags) photometry covering a relatively short timespan (2–8 hr times of ingress, mid-transit, and egress based on the RV fit in over five nights) from the Hereford Arizona Observatory, and Section 2.4 for the two methods of RV fitting (“C”isfore = 0, less precise (few percent), but more comprehensive photome- “E” is for non-zero eccentricity), and an assumed transit duration try consisting of 7194 epochs taken over roughly three years of 3.3 hr, corresponding to a nearly central transit. The widths No. 2, 2010 DISCOVERY OF TYC 2949-00557-1b 1191

Figure 3. Photometric observations with the Hereford Arizona Observatory telescope. The horizontal axis is elapsed time each night in hours. The star is photometrically stable with an rms of 2–4 mmag. The increased scatter towards the end of 2009 February 21, 2009 February 27, and most of 2010 April 15 are due to observing at high air mass. The vertical bars are the predicted times of ingress, mid-transit and egress based on the RV fit in Section 2.4 and an assumed transit duration of 3.3 hr. The two mid-transit estimates are based on the two results from the RV fitting (“C”isfore = 0, “E” is for the case where e is left as a free parameter). The widths correspond to the uncertainties in the mid-transit times. (A color version of this figure is available in the online journal.) correspond to the uncertainties in the mid-transit times. There is as the full MARVELS survey). TYC 2949-00557-1 is in one no evidence of a transit at the most likely depth (∼0.8%) and du- of KELT’s target fields and is a good example of this synergy. ration during these observations. In Section 3.3, we consider the We use the KELT photometry to characterize the photometric uncertainties in the ephemeris and properties of the primary, as variability of the host star, and to search for signatures of transits well as a range of impact parameters, to quantify the confidence of the companion. We first briefly describe the data reduction, with which we can exclude transits in this system. and then describe the light curve analysis. Images of the field are flat-fielded, and then relative pho- 3.2. Relative Photometry with KELT tometry is extracted using the ISIS image subtraction pack- KELT North survey is a wide-field photometric survey of age (Alard & Lupton 1998), in combination with DAOPHOT ∼40% of the northern sky designed to monitor fairly bright (Stetson 1987) to perform point-spread function fitting pho- (8

Figure 5. Small filled circles show the KELT-N light curve for TYC 2949- Figure 4. Top panel: KELT-N light curve for TYC 2949-00557-1. Bottom 00557-1 phased according to the ephemeris from the joint RV fit (see panel: Lomb–Scargle periodogram of the KELT data, showing no evidence for Section 2.4). The larger filled squares with error bars are binned 0.025 in phase. any significant periodicities for periods of P = 1–10 days, including the period of the RV companion (vertical dashed line) and the first harmonic (vertical detect them with a signal-to-noise ratio of dotted line).   (A color version of this figure is available in the online journal.) R 1/2 δ S/N ∼ N 1/2 (2) we see no evidence for photometric variability for this star (see πa σ below) we further scale the errors to force χ 2/dof = 1tobe where N = 7194 is the number of data points, and σ ∼ 3.5% conservative. TYC 2949-00557-1 happened to fall in an overlap ∼ 2 region of two target fields, and as a result we had two sets of light is the typical uncertainty. Thus, S/N 3(r/RJ ) , which is curves for the target. These data were reduced independently, marginal unless the companion has a radius significantly larger and then combined after subtracting the difference between their than . On the other hand, the majority of the Hereford weighted mean magnitudes. data is generally of sufficient quality to detect transits at the The KELT light curve for TYC 2949-00557-1 is shown in expected depth, and one night covers the predicted transit time Figure 4. It contains 7194 data points spanning 3.2 years, and for the companion. Unfortunately, there is no indication of a has a weighted rms of 3.5%. We search for variability using transit at the expected time. a weighted Lomb–Scargle periodogram with floating mean We nevertheless proceed with a quantitative search for a (Lomb 1976; Scargle 1982; Cumming et al. 1999), and find no transit signal. We combine the KELT data with the Hereford significant peaks (see Figure 4), and in particular no evidence data from 2010 April 15 after first subtracting the difference for periodic variability near the period of the companion between their weighted mean magnitudes. The slight difference (P 5.69 days), or the first harmonic (P/2). Figure 5 shows the in passband between the HAO and KELT data do not affect KELT lightcurve, phased according to the best-fit RV ephemeris the ability to detect a transit in the combined data set. We use (Table 4), as well as binned 0.025 in phase (3.46 hr). The rms the distribution of companion periods P and expected transit of the binned data is ∼2.4 mmag, and the χ 2/dof = 0.91 times Tc from the MCMC analysis of the combined RV data for a constant fit, indicating a low level of correlated noise, described in Section 2.4. For each combination of Tc and P (i.e., and no evidence for variability at the few mmag level. We for each link in the Markov Chain), we draw a random value limit the amplitude of any variability at P/2tobe2 mmag; for the Teff,logg, and [Fe/H] of the primary from a Gaussian unfortunately this is well above the level of ellipsoidal variability distribution, with the central values and dispersions given in expected for this companion of ∼0.03 sin i mmag (Pfahl et al. Table 5, as determined from the spectroscopic analysis described 2008). in Section 5.2. We then use the Torres et al. (2010) empirical relations to estimate the mass M and radius R of the primary 3.3. Limits on Transits for those values. We add an additional offset to M and R drawn from Gaussians with dispersions equal to the dispersions of the Given the relatively high a priori transit probability of fits to the empirical relations in Torres et al. (2010); specifically TYC 2949-00557-1b of ∼R/a ∼ 8%, where R is the stellar 6.4% in M and 3.2% in R. Finally, we draw a random value of radius, we searched for transits in the KELT dataset combined the impact parameter of the transit in units of the radius of the with the Hereford data from 2010 April 15. The expected transit star in the range [0, 1]. Assuming a radius for the companion, duration is ∼RP/(πa) ∼ 3.4 hr for a central transit, and the we then compute the expected transit curve using the routines ∼ 2 ∼ 2 expected fractional depth is δ (r/R) 0.8%(r/RJ ) , of Mandel & Agol (2002), using limb-darkening coefficients where r is the radius of the companion. The expected radius from Claret (2000), assuming that both the KELT and Hereford of the companion depends on its true mass, as well as the bandpasses roughly correspond to R. We then fit this curve to the ∼ 2 age of the system, but is likely to be 1 RJ (Baraffe et al. combined data set, and compute the improvement in χ relative 2003). Unfortunately, while the KELT data has excellent phase to a constant flux fit to the data. We repeat this for each link in coverage, it is not of sufficient quality to detect or rule out the the Markov chain, as well as for a variety of different companion expected signal. If transits were present, we should expect to radii. No. 2, 2010 DISCOVERY OF TYC 2949-00557-1b 1193

Table 5 Stellar Host Properties of TYC 2949-00557-1a

Parameter Value Uncertainty α (J2000)b 101.921152 (deg) 06:47:41.076 (HH:MM:SS) δ (J2000)b 42.009332 (deg) 42:00:33.60 (DD:MM:SS) B 12.846 0.023 V 12.142 0.031 Rc 11.750 0.039 Ic 11.391 0.043 J2MASS 10.820 0.022 H2MASS 10.474 0.021 K2MASS 10.421 0.018 Teff (K) 6135. 40. log (g[cm s−1]) 4.4 0.1 [Fe/H] 0.32 0.01 v sin I (km s−1)7.1. Mprimary (M) 1.25 0.09 Rprimary (R) 1.15 0.15 +66 d (pc) 413 −57

Notes. a Figure 6. Probability that transits of a companion are excluded at levels of BVRI magnitudes are unweighted averages from Table 6. Δ 2 ={ } b χ 9, 16, 25 based on the analysis of the combined Hereford and KELT Coordinates taken from the Tycho 2 Catalog (Høg et al. 2000) photometric data sets, as a function of the radius of the companion. The black, long-dashed line is a case where boxcar-shaped transits were used as a test, and is for Δχ 2 = 16. Transits of companions with radius r  0.75 R can be 2 J We search for significant improvements in χ whichwouldbe excluded at the 95% confidence level. 2 indicative of a detection. Our best-fit has Δχ =−11.7 relative (A color version of this figure is available in the online journal.) to a constant flux fit. In order to asses the significance of this improvement in χ 2, we repeat the search for “anti-transit,” i.e., Table 6 signals with the same shape as a transit but corresponding to an Measured Photometry for TYC 2949-00557-1 from Hereford Arizona increase in flux (see et al. 2006), and find improvements Observatory 2 in χ at similar levels. We therefore conclude that there is no Banda Flux Uncertainty evidence for a transit in the combined KELT and Hereford data. (mag) (mag) Given that we have not detected any evidence of transits, we B 12.839 0.023 now ask what the probability is that we would have detected a V 12.135 0.031 transit of a given radius, assuming that the companion transits Rc 11.750 0.039 (i.e., b  1). To do this, we simply determine what fraction Ic 11.390 0.043 of the steps in the Markov Chain described above result in an g 12.450 0.020 2 r 11.937 0.016 increase in χ above a certain level, as a function of the radius  of the companion. This result is shown in Figure 6,forΔχ 2 = i 11.811 0.028 {9, 16, 25}.TheΔχ 2 values were chosen as representative BSloan 12.853 0.023 V values: Δχ 2 = 16 is the likely detection limit, 25 is chosen Sloan 12.150 0.024 Rc 11.749 0.029 as a conservative limit, and 9 is chosen to straddle the true Sloan IcSloan 11.392 0.043 detection limit. The black, long-dashed line is a case where flat- a bottomed, boxcar-shaped transits (no ingress/egress) were used Note. Measurements labeled with Sloan are converted from the Sloan filter 2 and represents Δχ = 16. It shows that detailed modeling of observations using Smith et al. (2002) transformation equations. limb darkening and the ingress/egress has little effect on the final results of this test. Given the properties of the noise as estimated from the improvements in the fits from “antitransits,” stars and 15 SDSS standard stars were used as calibrators. The signals with Δχ 2  16 are likely to have been reliably detected. conversion equations of Smith et al. (2002) were used to con- vert the Sloan filter measurements into the BVRI system. The Thus, we can exclude transits of a companion with r  0.75 RJ . at the 95% confidence level. Baraffe et al. (2003) predict radii agreement between the observed BVRI magnitudes and those converted from Sloan magnitudes agree within the measure- of  0.77 RJ for brown dwarfs with m 60 MJ for ages of  5 Gyr. We conclude it is unlikely this companion transits, ment error. We adopt the unweighted average of each BVRI unless the system is substantially older than 5 Gyr, which is measurement and the larger of the two statistical errors for the unlikely given the effective temperature and surface of final magnitude results. Table 6 summarizes the measured mag- the host star (Section 6). nitudes and final results for the multi-band photometry. Near-IR fluxes are taken from the 2MASS (Skrutskie et al. 2006) Point Source Catalog and are presented in Table 5. 4. ABSOLUTE PHOTOMETRY The Tycho-2 catalog’s (Høg et al. 2000) V-band magnitude 5. CHARACTERIZATION OF THE HOST STAR for this object is 11.840, however, Tycho magnitudes are known 5.1. SED Analysis to significantly degrade beyond VT > 11. Measurements were taken at Hereford Arizona Observatory using both B,V,Rc,Ic We use the BVRI fluxes along with the 2MASS near-IR and Sloan g,r,i filter sets. A total of 64 Landolt standard data to fit model spectral energy diagrams (SEDs) to derive 1194 FLEMING ET AL. Vol. 718

Figure 8. HET/HRS template spectrum in black with best-fit model in red for the Mgb region. The spectra have been continuum normalized and the relative flux density is plotted against the wavelength in Angstroms. (A color version of this figure is available in the online journal.)

ing to the prescription of Barklem & O’Mara (1998). The line lists used are drawn from a variety of sources. Updated atomic lines are taken mainly from the Vienna Atomic Line Database (VALD) database (Kupka et al. 1999). The molecular species 2 Figure 7. Top: χ map in Teff − AV space showing the degeneracy between CH, CN, OH, CaH, and TiO are provided by B. Plez (see Plez & extinction and Teff . Inner contours are 1σ uncertainties for Teff and AV assuming Cohen 2005), while the NH, MgH, and C2 molecules are from they are the only significant degrees of freedom, outer contours assume all four the Kurucz linelists. The solar abundances used here are the parameters are significant. Bottom: NextGen model overplotted on the observed i ii fluxes. The dotted curve is a blackbody SED for the best-fit T . same as Asplund (2005). The Fe and Fe lines used for the line eff analysis were compiled by Santos et al. (2004), who use solar (A color version of this figure is available in the online journal.) Fe abundance to derive gf values. Both Asplund (2005) and San- approximate stellar parameters. NextGen models from tos et al. (2004) use HARPS solar spectra and an Fe abundance Hauschildt et al. (1999) in grids of 100 K for T ,0.5dex of 7.45. eff = ± = ± for log g and 0.5 dex for the , represented by We derive a Teff 6135 40 K and [Fe/H] 0.32 0.01, i = ± [Z/H], are fit along with the line-of-sight extinction. There based on Fe excitation equilibrium and a log g 4.4 0.1 i ii is a degeneracy between line-of-sight extinction and derived based on the ionization equilibrium of Fe and Fe lines and by fitting the wings of the Mgb triplet at 5167, 5172 and 5183 Å. Teff when fitting SEDs with unknown extinction. Figure 7 2 The error estimates are based on the equivalent width of Fe (top) shows the χ map in Teff − AV space. The interior con- tours represent 1σ uncertainties assuming that T and A are lines and the errors of Fe abundances from the individual lines. eff V = −1 the only two free parameters. The exterior contours are the A value ξt 1.65 km s is derived by forcing weak and strong Fe i lines to give the same abundances. Fitting 1σ uncertainties with all four parameters (Teff, AV ,logg and the Fe lines in the Mgb region yields a rotational velocity of [Z/H]) as free parameters. We find the global minimum in −1 2 v sin I = 7kms . We only used the 110 Fe i lines weaker χ -space is for a Teff = 6000 K, log g of 5.0, [Z/H] of 0.0 than 100 mÅ for the analysis. Figure 8 shows the continuum and extinction AV = 0.5. Figure 7 (bottom) shows the best-fit NextGen model along with the observed fluxes. Other solutions normalized spectra in black and the best-fit model in red for the Mgb region. exist within the 1σ contours at cooler Teff and smaller AV ;how- In addition to the HET templates, we obtained high-resolution ever, the dust maps of Schlegel et al. (1998)giveanAV of 0.45 (R ∼ 31,000) Echelle spectra using the ARCES instrument along this line of sight, consistent with the best-fit AV of 0.5 assuming that the star is behind the majority of the dust along (Wang et al. 2003) on the APO 3.5m telescope. Seven exposures this line of sight. The hotter temperature solution is also con- for a total of 63 minutes of integration were obtained. Data was reduced using a modified IRAF script originally written sistent with the Teff derived from Echelle spectra presented in Section 5.2. by J. Barentine and J. Krzesinski for ARCES data. Spectra are corrected for bias and dark subtraction, cosmic rays and 5.2. Spectral Synthesis bad pixels. Flatfielding is performed using a quartz lamp and two different sets of integration times: a “blue” set of 4 min In order to derive physical properties of the host star and integrations using a blue filter and a “red” set of 7 s integrations estimate the minimum mass of the companion, stellar tem- with no filter in the beam. These sets are then combined to form plates from the HET observations that do not contain iodine a master flatfield image. The two different quartz sets are used to lines were used to derive parameters of the host star. We maximize the signal-to-noise in both the blue and red end of the use the latest MARCS model (Gustafsson et al. spectrum. Spectra are wavelength calibrated using a sequence 2008) for the analysis. Generation of synthetic spectra and of 10-s ThAr integrations taken a few times during each night. the line analysis were performed using the turbospectrum code The star HD 42088 (spectral type 06V) was observed to remove (Alvarez & Plez 1998), which employs line broadening accord- telluric lines. No. 2, 2010 DISCOVERY OF TYC 2949-00557-1b 1195

Figure 9. Portion of the ARCES spectrum (R ∼ 31,000) used in the SME analysis. The input spectrum is in white with the best-fit model overlaid in black.

The spectrum was analyzed using the IDL-based program Spectroscopy Made Easy, or SME (Valenti & Piskunov 1996). This code uses synthetic spectra and multidimensional least- squares minimization to determine the best set of stellar pa- rameters for an observed spectrum. These parameters include effective temperature, , metallicity, microturbu- lence, macroturbulence, projected rotational velocity, and the RV. We follow the guidelines from previous spectroscopic stud- ies of host stars that used SME (e.g., Valenti & Fischer 2005; Stempels et al. 2007). We used three-dimensional interpola- tion on the Kurucz (1993) grid of local thermodynamic equi- librium (LTE) model atmospheres and the VALD database to obtain line data for transitions with predicted absorption cores deeper than 0.5% of the continuum. For the VALD queries, Figure 10. HR diagram as a function of Teff and log g based on Yonsei-Yale we used solar abundances, Teff = 5770 K, log g = 4.44, and models. The solid track is for the best-fit stellar parameters −1 of 1.25 M and [Fe/H] = +0.32. The two dashed tracks are for masses of ξt = 0.866 km s . 1.25 ± 0.09 M and represent the 1σ uncertainties on the mass. Blue dots are For the SME analysis, as suggested by Stempels et al. (2007), { } −1 the location of the star at ages of 1,2,3 Gyr, respectively. TYC 2949-00557-1 we fixed the parameter ξt to 0.85 km s , in order to decouple the is consistent with a ZAMS star that has an age no older than ∼ 2Gyr. correlation between microturbulence ξt and metallicity. For the (A color version of this figure is available in the online journal.) macroturbulence ζt , we follow the empirical relation of Valenti −1 & Fischer (2005) which gives ζt = 4.5kms for a star with Teff ∼ 6200 K. We were unable to obtain consistent results from the gravity-sensitive Mgb triplet region, therefore we fix log g at systems. We use this empirical relation to derive the mass and three values of 4.3, 4.4, and 4.5, corresponding to the 1σ range radius of the primary using the values of Teff,logg and [Fe/H] obtained from the HET template spectra. Applying the equations determined from the HET spectra. We set as free parameters Teff, for log M and log R yields a mass of M = 1.25 ± 0.09 M and [M/H], v sin I, and the RV vrad and utilize the metal-rich region = ± of 6000–6200 Å. The uncertainties of the parameters are derived a radius of R 1.15 0.15 R. Correlations of the best-fit from the range of best-fit results using the three fixed log g coefficients from Torres et al. (2010) are included in the errors, = +27 = +0.009 but correlations of Teff,logg and [Fe/H] are not considered. The values. We derive Teff 6246−45 K, [M/H] 0.3615−0.027 = ± −1 reported scatter in the relation as found in Torres et al. (2010) and v sin I 7 1kms , in reasonable agreement with the = = values derived from the HET spectra. Figure 9 shows the best-fit of σlogm 0.027 and σlogr 0.014 are also included in the model in black and the input spectrum in white for a portion of mass and radius uncertainties, respectively, by adding them in the 6000–6200 Å range used in the fitting. quadrature. The SME analysis of the ARCES spectrum is used as an Figure 10 compares the spectroscopically measured Teff and log g of TYC 2949-00557-1 (red error bars) to a theoretical stel- independent check on the derived temperature and metallicity 2 from the HET spectra. Because it is based on a smaller lar evolutionary track from the Yonsei-Yale (“Y ”) model grid wavelength region, and cannot be used to independently derive (see Demarque et al. 2004, and references therein). The solid curve represents the evolution of a single star of mass 1.25 M the surface gravity, we chose to adopt the results from the HET = analysis for the final stellar properties. and metallicity of [Fe/H] +0.32, starting from the zero-age main sequence (lower left corner), across the Hertzsprung gap, 6. DETERMINATION OF HOST STAR MASS and to the base of the branch. Symbols indicate various AND RADIUS ages along the track labeled in Gyr. The dashed curves repre- sent the same evolutionary track but for masses ±0.09 M, An alternative to interpolating isochrone models to determine representative of the 1σ uncertainty in the mass from the stellar properties is to apply the analytical equations derived by Torres et al. (2010) relation. The filled gray region between the Torres et al. (2010) using measurements of eclipsing binary two mass tracks represents the range of expected locations of a 1196 FLEMING ET AL. Vol. 718 star like TYC 2949-00557-1 given the 1σ mass uncertainty and measured metallicity. We emphasize that we have not directly measured the mass of TYC 2949 but rather derived the mass using the empirical relation of Torres et al. (2010). Our pur- pose here is not to test the accuracy of the theoretical stellar evolutionary tracks, but rather to constrain the evolutionary sta- tus of the TYC 2949 system. The spectroscopically measured Teff,logg, and [Fe/H] place TYC 2949 near the zero-age main sequence, with an age of at most ∼2Gyr. The distance to the host star can be computed once the bolometric luminosity is known. We use the Stefan–Boltzmann law to derive the luminosity of the star using the Teff found in Section 5.2 and the radius calculated by the Torres et al. (2010) relation. The absolute V magnitude is then given by

=− L − MV 2.5log + MBol BCV (3) L where MBol is the bolometric of the , here assumed to be 4.74, and BCV is the bolometric correction to the V band. We adopt a value of BCV =−0.175, interpolated from Table 15.7 in Drilling & Landolt (2004) based on the Figure 11. Cumulative probability that the mass of the companion to TYC 2949- 00557-1 is less than a given mass in solar masses. These probabilities account Teff of TYC 2949-00557-1. The distance can then be calculated for the uncertainties and covariances between the parameters of the RV fit, via the distance modulus and assuming a value for the line- the uncertainty in the mass of the primary, assuming a uniform distribution of of-sight extinction. If we assume AV = 0.45 ± 0.1, which is cos i, and adopting various priors for the distribution of companion mass ratios consistent with both the Schlegel et al. (1998) dust maps and dN/d log q. the best-fit value obtained from the SED fitting in Section 5.1, +66 we derive a distance to the system of 413 −57 pc. This distance is consistent with our implicit assumption that the star is behind value for the spectroscopically determined primary parameters the majority of the dust along this line of sight, for likely values T ,logg, and [Fe/H], according to a Gaussian distribution of the thickness of the dust layer. The quoted uncertainty in eff centered on the best-fit values and with dispersions equal to the the distance includes the uncertainties in the stellar radius, uncertainties (given in Table 5). We use these values of T , effective temperature, line-of-sight extinction and apparent eff log g, and [Fe/H] to estimate the primary mass M using the V-band magnitude. empirical relation of Torres et al. (2010). We add an additional offset to M drawn from a Gaussian with dispersion equal to the 7. COMPANION MASS dispersion in the fit to this empirical relation (6.4%). We draw a Given the orbital parameters from the joint RV fit random value of cos i from a uniform distribution in the range (Section 2.4), and the estimate of the mass of the host star de- (0, 1), and then solve for the mass m of the secondary (note we rived from the spectroscopic parameters (Section 5.2), we can do not assume that m  M). estimate a minimum mass for the companion of We weight the resulting distribution of m by a prior on the luminosity ratio l and a prior on the mass ratio q. Specifically,  mmin = 64.3 ± 3.0 MJ , (4) we assume luminosity ratios of l 0.1 are excluded by the lack of features due to the companion in the high-resolution spectra. where we have assumed a . This minimum mass We assume a flux ratio relationship of the form l ∝ q4.5,asis is based only on the RV and stellar parameters, and ignores roughly appropriate for main-sequence stars. The exponent of the fact that edge-on configurations are likely excluded given 4.5 is derived by fitting the values found in Table 1 of Torres the lack of transit signature. Since the minimum mass is below et al. (2010) for stars with 0.5

Matzner, C. D., & Levin, Y. 2005, ApJ, 628, 817 Siverd, R. J., Pepper, J., Stanek, K., Pogge, R. W., Gaudi, B. S., & DePoy, D. Mazeh, T. 2008, in , Observational Evidence for Tidal Interactions in Close L. 2009, in IAU Symp. 253, Transiting Planets, ed. F. Pont, D. Queloz, & D. Binary Systems, ed. M.-J. Groupil & J.-P. Zahn (EAS Publications Ser. 29; Sasselov (Cambridge: Cambridge Univ. Press), 350 Paris: EDP Sciences), 1 Skrutskie, M. F., et al. 2006, AJ, 131, 1163 Mazeh, T., Goldberg, D., Duquennoy, A., & Mayor, M. 1992, ApJ, 401, 265 Smith, J. A., et al. 2002, AJ, 123, 2121 Metchev, S. A., & Hillenbrand, L. A. 2009, ApJS, 181, 62 Stassun, K. G., Mathieu, R. D., & Valenti, J. A. 2006, Nature, 440, 311 Niedzielski, A., Nowak, G., Adamow,´ M., & Wolszczan, A. 2009, ApJ, 707, Stassun, K. G., Mathieu, R. D., & Valenti, J. A. 2007, ApJ, 664, 1154 768 Stempels, H. C., Collier Cameron, A., Hebb, L., Smalley, B., & Frandsen, S. Omiya, M., et al. 2009, PASJ, 61, 825 2007, MNRAS, 379, 773 Patel, S. G., Vogt, S. S., Marcy, G. W., Johnson, J. A., Fischer, D. A., Wright, J. Stetson, P. B. 1987, PASP, 99, 191 T., & Butler, R. P. 2007, ApJ, 665, 744 Torres, G., Andersen, J., & Gimenez,´ A. 2010, A&AR, 18, 67 Pfahl, E., Arras, P., & Paxton, B. 2008, ApJ, 679, 783 Tull, R. G. 1998, Proc. SPIE, 3355, 387 Pepper, J., Gould, A., & Depoy, D. L. 2003, Acta Astron., 53, 213 Udry, S., Mayor, M., Naef, D., Pepe, F., Queloz, D., Santos, N. C., & Burnet, Pepper, J., et al. 2007, PASP, 119, 923 M. 2002, A&A, 390, 267 Plez, B., & Cohen, J. G. 2005, A&A, 434, 1117 Valenti, J., & Fischer, D. 2005, ApJS, 159, 141 Pont, F. 2009, MNRAS, 396, 1789 Valenti, J., & Piskunov, N. 1996, A&AS, 118, 595 Ramsey, L. W., et al. 1998, Proc. SPIE, 3352, 34 van Eyken, J. C., Ge, J., Mahadevan, S., & DeWitt, C. 2004, ApJ, 600, 79 Santos, N. C., Israelian, G., & Mayor, M. 2004, A&A, 415, 1153 van Eyken, J. C., Ge, J., & Mahadevan, S. 2010, ApJS, in press Santos, N. C., et al. 2002, A&A, 392, 215 (arXiv:1005.5564v1) Scargle, J. D. 1982, ApJ, 263, 835 Wang, S., et al. 2003, Proc. SPIE, 4841, 1145 Schlegel, D. J., Finkbeiner, D. P., & Davis, M. 1998, ApJ, 500, 525 Wittenmyer, R. A., Endl, M., Cochran, W. D., Ram´ırez, I., Reffert, S., Shetrone, M., et al. 2007, PASP, 119, 556 MacQueen, P. J., & Shetrone, M. 2009, AJ, 137, 3529